Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/472756
Title: | Hybrid features based signboard text recognition system for Gurmukhi script |
Researcher: | Bains, Jasleen Kaur |
Guide(s): | Sharma, Anuj |
Keywords: | Deep learning Gurmukhi Script Indic script recognition Signboard Text Recognition Word recognition |
University: | Panjab University |
Completed Date: | 2022 |
Abstract: | Signboard text recognition has gained significant attention in recent decades. This study proposes a signboard text recognition system for the Gurmukhi script and examines the state-of-the-art work for signboard text recognition in various scripts. The study focuses on the computation of correct features to represent data efficiently and achieve high accuracy in text recognition systems. The proposed system recognizes Gurmukhi signboard image strokes using dynamic, static, and hybrid feature sets. The offline text lacks dynamic information on the writing order or nature of the trajectories of stroke. A recovery of the drawing order technique has been used to retrieve the trajectory of a stroke, aiding in computing a dynamic feature vector based on chain codes or trajectory points for text recognition. Stroke recognition has been performed using Conv1D, SVM, and HMM classifiers for dynamic feature alone. The best overall recognition accuracy using a hybrid feature set has been achieved using the SVM and Conv1D deep learning method as 91.37% and 93.39%. The character and word formation processes of Gurmukhi words in signboard images have been performed. The Gurmukhi strokes are categorized into major-dependent and dependent strokes. The rearrangement of strokes is performed to form a complete character during the character formation process. For word formation, an understanding of the Gurmukhi script has been used to define the process of word formation. The Gurmukhi words having one or more zones can be formed using the proposed word formation process. The proposed word formation process achieved an overall word recognition accuracy of 82.12% using the SVM and 83.86% using the Conv1D deep learning method on 1000 signboard word images. The techniques used in this study can be used in real-life applications such as signature verification, and document recognition, and can be extended to word recognition of other Indic scripts such as Devanagari. newline |
Pagination: | Bibliography 197-208p. |
URI: | http://hdl.handle.net/10603/472756 |
Appears in Departments: | Department of Computer Science and Application |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_titlepage.pdf | Attached File | 28.71 kB | Adobe PDF | View/Open |
02_prelim pages.pdf | 1.45 MB | Adobe PDF | View/Open | |
03_chapter_1.pdf | 435.03 kB | Adobe PDF | View/Open | |
04_chapter_2.pdf | 212.33 kB | Adobe PDF | View/Open | |
05_chapter_3.pdf | 2.2 MB | Adobe PDF | View/Open | |
06_chapter_4.pdf | 1.21 MB | Adobe PDF | View/Open | |
07_chapter_5.pdf | 1.22 MB | Adobe PDF | View/Open | |
08_chapter_6.pdf | 779.61 kB | Adobe PDF | View/Open | |
09_chapter_7.pdf | 1.85 MB | Adobe PDF | View/Open | |
10_chapter_8.pdf | 274.85 kB | Adobe PDF | View/Open | |
11_annexure.pdf | 312.26 kB | Adobe PDF | View/Open | |
80_recommendation.pdf | 303.77 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: